Photodetectors capable of detecting a single photon (i.e., a single “particle” of optical energy) have enabled systems directed toward applications such as single-photon detection and low-light-level imaging. Due to its performance, reliability, cost, and ease of deployment, the semiconductor-based single-photon avalanche diode (SPAD) has been the basis for most of these systems. In recent years, single-photon infrared applications have become of particular interest. As a result, SPADs formed in the indium-gallium-arsenide/indium-phosphide material systems have been a focus of much research and development. Although there has been significant improvement in this device technology over the past several years, single photon receivers have generally been limited to operational rates (i.e., counting rates) below 10 MHz. More recently, however, there has been increased interest in defense applications that require counting rates in the GHz range, such as ultra-secure quantum cryptography systems, quantum information processing, quantum computing, and long distance free-space optical communications. Unfortunately, this combination of high operation rate and wavelength range has historically been difficult, if not impossible, to achieve.
An avalanche photodiode derives its name from the manner in which its output signal is created. When photons are absorbed by an avalanche photodiode, their energy frees bound charge carriers (electrons and holes) that then become free-carrier pairs. In the presence of an electric field (due to a bias voltage applied to the photodiode), these free-carriers are accelerated through a region of the avalanche photodiode referred to as the “multiplication region.” As the free carriers travel through the multiplication region, they collide with other carriers bound in the atomic lattice of the semiconductor, thereby generating more free carriers through a process called “impact ionization.” These new free-carriers also become accelerated by the applied electric field and generate yet more free-carriers. This avalanche event can occur very rapidly and efficiently and it is possible to generate several hundred million free-carriers from a single absorbed photon in less than one nanosecond.
In order to enable detection of a single photon, a SPAD is biased with a reverse bias voltage having a magnitude greater than the magnitude of its “breakdown voltage,” which is the bias level above which free-carrier generation can become self-sustaining and result in run-away avalanche. This is referred to as “arming” the device. When the SPAD is armed, a single free carrier pair created by the absorption of a single photon can create a runaway avalanche resulting in an easily detectable macroscopic current.
Once a photon has been detected, the self-sustained avalanche must be stopped before the arrival of another photon can be detected. Referred to as “quenching,” the avalanche is stopped by reducing the magnitude of the applied reverse bias voltage below the magnitude of the breakdown voltage.
In typical operation, a periodic bias signal (referred to as a “gating signal”) is applied to a SPAD to arm and quench the device during each bit period. The maximum frequency at which a SPAD can be gated is primarily determined by how fast an avalanche event can be stopped once it is detected and how fast the SPAD can be re-armed once the avalanche event has been quenched.
Unfortunately, the frequency of the gating signal is limited by the fact that not all free carriers are instantaneously swept out of the avalanche region once the device is quenched. With each avalanche event, some fraction of the electrical carriers created will become trapped at defects (e.g., crystalline defects, impurities, etc.) in the multiplication region. These trapped carriers become released in a temporally random manner as a function of temperature, the type of trap state, and applied bias voltage. As a result, the trapped carrier population decays exponentially with time.
Detrapping can occur without consequence while the SPAD remains in its quenched state. If a trapped carrier is released after the SPAD has been re-armed, however, it is likely to trigger an avalanche event that is indistinguishable from one caused by absorption of a photon. Avalanche events induced by carriers created by any mechanism other than photo-excitation (i.e., in the absence of input photons) are referred to as “dark counts,” and dark counts caused by the detrapping of trapped carriers in the multiplication region are referred to as “afterpulses.” The probability of an afterpulse decays exponentially with the hold-off time between quenching a SPAD and re-arming it, so decreasing the hold-off time leads to a higher probability of afterpulsing.
Afterpulsing represents one of the primary roadblocks to high-rate photon counting. Prior-art methods for reducing afterpulsing have been focused on limiting the charge flow induced by an avalanche event, thereby limiting the number of carriers that can potentially become trapped. One such method relies on the use of a negative-feedback element monolithically integrated with the SPAD to form a negative-feedback avalanche diode (NFAD), such as is described in U.S. Pat. No. 7,719,029 entitled “Negative Feedback Avalanche Photodiode,” issued May 18, 2010, and which is incorporated herein by reference. Unfortunately, limiting the charge flow alone leads to additional complications since it results in smaller amplitude signals to be detected. These small signals can be masked by the large transient signals that are generated by the interaction of high-frequency gating signal components with reactance inherent in the SPAD structure. As a result, approaches for high-frequency transient cancellation have been developed in an effort to enable accurate detection of small-amplitude avalanche pulses.
Ideally, a gating signal maximizes the portion of each bit period during which a SPAD is armed, while providing a hold-off time just sufficient to minimize afterpulsing. Further, it is highly desirable to provide gating signals that have fast transitions (i.e., sub-nanosecond) between their low state (quenching bias) and high state (arming bias). Unfortunately, these fast transitions represent high frequency components in the gating signal that serve to generate large capacitive transients in the SPAD. These transients can couple into the SPAD output signal thereby reducing its signal-to-noise ratio. As a result, methods for suppressing these capacitive transients to enable more accurate detection of the typically much smaller signals induced by SPAD avalanche events have been developed. These methods fall into one of two general approaches, after-the-fact cancellation of generated capacitive transients or avoidance of the generation of the capacitive transients in the first place.
One of the more promising methods for avoiding generation of capacitive transients relies on the use of a gating signal devoid of high frequency components—in particular, a sinusoidal gating signal. An example of such a method is described in U.S. Pat. No. 7,705,284, issued Apr. 27, 2010, which is incorporated herein by reference. In this method, a gating signal comprising a D.C. voltage and a substantially pure sinusoidal gating signal is applied to a SPAD. Because the sinusoidal gating signal contains essentially only one frequency component, gating signal components in the SPAD output signal are concentrated at this frequency component and its harmonics. The signal components that arise from the reception of single photons in the SPAD output signal are akin to an impulse response, however. As a result, these photon signal components include frequencies that are widely spread across the frequency spectrum. Relatively straightforward electrical filters can be applied to the SPAD output signal, therefore, to remove the gating signal components and facilitate detection of avalanche events due to reception of a single photons—even when the avalanche amplitude is small. Sinusoidal-gating methods have demonstrated afterpulsing as low as 3.4% with a photon-detection efficiency of approximately 10.5% for SPADs biased with gate signals having a frequency of 2 GHz.
Unfortunately, while the sinusoidal-gating concept simplifies elimination of the gating-signal components, the use of the “top” portion of the sine wave as the SPAD gating bias means that the excess bias changes dramatically throughout a significant fraction of the gate duration. This is due to the fairly shallow slope of the rise and fall of the gate imposed by the simple sine-wave functional form. As a result of the changing excess bias, the photon detection efficiency changes as well. If photon arrivals can be aligned to the relatively “flat” portion of the gate with very low jitter, then the shallow rise and fall times will not be a serious issue. However, for GHz-rate gating with effective gate widths on the order of just 100-200 picosecond (ps), jitter in the photon arrival time of just 50-100 ps can significantly impact the effective photon detection efficiency from one count to the next. As applications demand higher operating frequency, this problem will be exacerbated.
The use of a “squarer” gate signal having more rapid rise and fall times provides a relatively wider portion of bit period in which the bias is “flat.” As a result, the photon detection efficiency is relatively constant. Further, a more sharply falling edge of a gating signal provides more rapid avalanche quenching than a smooth sine wave. This results in less charge flow per avalanche, and consequently to a reduction in afterpulsing relative to sinusoidal gating.
As opposed to sinusoidal-gating methods, therefore, some prior-art gating methods derive some of the advantages of faster rise and fall time gate signals by employing square-wave gating signals and addressing the inevitable capacitive transients by canceling them out with additional circuitry. To date, some of the most promising results for transient cancellation have been obtained by employing a “self-differencing” circuit, as described by Yuan, et al., in “Multi-gigahertz operation of photon counting InGaAs avalanche photodiodes,” Applied Physics Letters, Vol. 96, 071101 (2010), and “High-speed single photon detection in the near infrared,” Applied Physics Letters, Vol. 91, 041114 (2007), each of which is incorporated herein by reference. In such methods, a square-wave gating signal is applied to a SPAD and the output signal of the SPAD is provided to a 50:50 splitter. The splitter splits the SPAD output signal into two signals, one of which is delayed by exactly one bit period of the gating signal. The delayed signal of a first bit period is then subtracted from the non-delayed signal from the next bit period. As a result, identical capacitive transients produced during sequential gate periods are canceled, leaving only any net avalanche signal that might occur. Self-differencing circuit methods have demonstrated afterpulsing as low as 1.4% with a photon-detection efficiency of approximately 11.8% for SPADs biased with gate signals having a frequency of 2 GHz.
An alternative approach to transient cancellation, after-the-fact, is based upon the use of a matched SPAD pair. The output from each SPAD is provided to a circuit element that provides an output signal based only on the difference between the two SPAD outputs, thereby by substantially rejecting capacitive transients included in both signals. An example of such an approach is described in “Balanced, gated-mode photon detector for quantum-bit discrimination at 1550 nm,” Optics Letters, Vol. 27, pp. 1827-1829 (2002).
Unfortunately, prior-art SPAD-gating methods have significant drawbacks that have limited their utility in practical single-photon detection systems. These drawbacks include: a need to operation at only a single fixed frequency, which limits their utility in communications applications, among other systems; high residual afterpulsing levels, which limits their signal-to-noise ratio and operating rates; complicated optical packaging; and relatively large form factors, which preclude easy deployment and commercialization.